• I recently joined NVIDIA as a Distinguished Engineer, where I’m developing VLSI AI Agents that leverage large language models (LLMs) to accelerate hardware design.
• My research interests include deep learning and information theory, Google Scholar.
• Past work experiences
  • I was a ML researcher at Google Research (see selected research projects below)

Recent News

• May 2026: Researchers extended our earlier WDR91 work using FEP and machine learning to achieve stronger potency, a further step toward a WDR91 drug candidate, in An Integrated Workflow Comprising AI, Physics and Experiment: Discovery of Nanomolar-Potent Inhibitors.
• April 2026: Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence, a multimodal model I contributed to, was released by NVIDIA.

Selected Research Projects

• Developing General AI Agents with Gemini Multimodality
  • Leveraged Langfun and Gemini’s multimodal capabilities to build a general-purpose AI agent, achieving SOTA performance on the GAIA benchmark in December 2024. This project demonstrated the effectiveness of Langfun + Gemini for complex reasoning and task completion across diverse domains.
  • Adapted the general-purpose AI agent idea to create a specialized vision-only AI agent for automated proofreading of large-scale 3D neuron reconstructions in mouse brain microscopy datasets on neuroglancer.
• Applying Deep Learning to Drug Discovery
  • Our publication in the Journal of Medicinal Chemistry, for the first time, discovered a novel small molecule ligand for WDR91 by using affinity-mediated DNA-encoded chemical library selection followed by deep learning. WDR’s unique β-propeller structure makes them attractive targets, but most human WDRs are unexplored compared to other major drug target families. The discovery of a drug-like small molecule and its covalent analog compounds will soon enable researchers to identify a WDR91 drug candidate.
  • Talk
• Studying secure communication from information theoretic perspective
  • Our publication in IEEE Transactions on Information Theory provided new insights into secure communication over broadcast channels by developing both inner and outer bounds for the rate equivocation region.
  • Ph.D. Dissertation

Education

• Ph.D. in Electrical and Computer Engineering, Syracuse University, 2009
• M.S. in Computer Science, Chinese Academy of Science, Institute of Automation, 2005
• B.S. in Electronic Engineering and Information Science, University of Science and Technology of China, 2002

Tools

• JAX implementation of MiniGPT
• PubChem and Mols2Grid

Tech Sharing

• Applying Deep Learning to Drug Discovery
• LLM as a judge